What Is the Purpose of the Google Agent Skills Repository?

The Google Agent Skills repository is a curated collection of skill definitions that enable AI agents to interact with Google products and services in a consistent, reusable way.

The Google Agent Skills repository serves as the foundational knowledge graph for agentic automation on Google Cloud. Each skill is a self-contained Markdown file—often with a front-matter block—that describes a specific capability, such as provisioning a GKE cluster, querying BigQuery, or configuring Workload Identity. The repository provides the commands, client-library snippets, and best-practice guidance an agent needs to execute tasks safely without custom code for each service.

Core Architectural Goals of Google Agent Skills

The Google Agent Skills repository is designed around three interconnected goals that standardize how AI agents discover and execute Google Cloud capabilities.

Standardized Agent Interaction

Every skill follows a uniform YAML/Markdown schema with fields like name, metadata, description, and structured reference sections. This standardization allows agents to programmatically discover, invoke, and reason about capabilities across the entire Google Cloud ecosystem.

In skills/cloud/gke-basics/SKILL.md, for example, the metadata block defines the skill's identity, parameters, and execution context:

name: gke-basics
description: Provision and manage GKE clusters using best practices
parameters:
  - name: cluster_name
    type: string
    required: true
  - name: region
    type: string
    default: us-central1

This schema enables any compliant agent to parse the skill without service-specific parsing logic.

Reusable Knowledge Base

The repository centralizes up-to-date, community-maintained recipes that eliminate duplication across separate projects. Each skill includes:

• CLI examples – Exact gcloud commands with required flags
• Client-library snippets – Python, Go, Java, Node.js implementations
• IaC guidance – Terraform or Config Connector configurations
• Security best practices – Workload Identity, private clusters, least-privilege IAM

The README.md at the repository root documents the full skill catalog and contribution guidelines, ensuring agents always act on current Google-recommended practices.

Plug-and-Play Integration for Agent Harnesses

The plugins/ directory bundles Google-product plugins that expose skills through a standardized plugin interface. Developers can add "Google-aware" abilities to their agents with minimal configuration.

Plugin	Framework	Installation
Claude	Anthropic Claude Desktop	npx skills add google/skills
Codex	OpenAI Codex CLI	codex skills install google/skills
Antigravity	Antigravity CLI	antigravity skills add google/skills

As documented in README.md#plugins, these plugins translate the skill schema into each framework's native tool-calling format.

Skill Structure and Discovery

Each skill file follows a predictable path pattern: skills/{category}/{skill-name}/SKILL.md. The repository organizes skills by domain:

• skills/cloud/ – Core Google Cloud services (Compute, GKE, Cloud Run)
• skills/data/ – BigQuery, Dataflow, Pub/Sub
• skills/security/ – IAM, Workload Identity, Cloud KMS
• skills/ai/ – Vertex AI, model deployment, MLOps pipelines

The front-matter schema enables automated indexing. Agents can scan the repository, parse YAML headers, and build a capability graph without human curation.

Practical Usage Examples

Installing a Skill via CLI

npx skills add google/skills

# Prompts selection: gke-basics, bigquery-query, workload-identity, etc.

This command is documented in README.md#installation.

Using GKE Basics in a Python Agent

from google.skills import gke_basics

# Provision an Autopilot cluster with security hardening

gke_basics.create_cluster(
    name="demo-cluster",
    region="us-central1",
    autopilot=True,
    private_nodes=True,
    master_authorized_networks=["10.0.0.0/24"]
)

The Python client patterns are described in skills/cloud/gke-basics/references/client-library-usage.md.

Executing via Native gcloud

gcloud container clusters create-auto demo-cluster \
  --region=us-central1 \
  --enable-private-nodes \
  --enable-master-authorized-networks \
  --master-authorized-networks=10.0.0.0/24

Flag-level details appear in skills/cloud/gke-basics/references/cli-reference.md.

Key Files in the Google Agent Skills Repository

File Path	Purpose
README.md	Repository overview, installation, skill catalog index
skills/cloud/gke-basics/SKILL.md	Canonical skill definition example
skills/**/SKILL.md	Skill instances following standardized schema
plugins/	Framework integration manifests (Claude, Codex, Antigravity)
LICENSE	Apache 2.0 open-source license

These files demonstrate how the Google Agent Skills repository transforms static documentation into machine-actionable capabilities.

Summary

• Google Agent Skills provides a standardized, machine-readable format for encoding Google Cloud expertise
• Each skill definition combines metadata, CLI commands, client-library code, and security guidance in one Markdown file
• The uniform schema enables automatic discovery and invocation by diverse agent frameworks
• Plugin integrations lower the barrier for developers to add Google Cloud capabilities to existing agents
• The community-maintained repository ensures agents operate on current best practices without fragmented, outdated code

Frequently Asked Questions

What format do Google Agent Skills use?

Skills are self-contained Markdown files with YAML front matter. The front matter defines name, metadata, parameters, and description fields, while the body contains narrative explanation, CLI examples, and code snippets in multiple languages. This format balances human readability with machine parseability.

How do agents discover available skills?

Agents scan the repository file structure—specifically skills/**/SKILL.md paths—and parse the YAML front matter of each file. The consistent schema allows agents to build a capability catalog without custom parsers for individual services, as implemented in the plugin handlers for Claude and Codex.

Can I contribute a new skill to the repository?

Yes. The repository accepts community contributions under the Apache 2.0 license. New skills must follow the established schema documented in README.md, include working code examples, and adhere to Google Cloud security best practices. Pull requests are reviewed for technical accuracy and consistency with existing skill patterns.

What's the difference between a skill and a plugin?

A skill is a domain-specific capability definition (e.g., "provision GKE cluster") stored in skills/**/SKILL.md. A plugin is a framework adapter in plugins/ that translates the skill schema into a specific agent runtime's tool-calling protocol. Skills are content; plugins are integration mechanisms.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}